Abstract: A neural network system is provided, implementing a generative model for autoregressively generating a distribution for a plurality of current filter-bank samples of an audio signal, wherein the current samples correspond to a current time slot, and each current sample corresponds to a channel of the filter-bank. The system includes a hierarchy of a plurality of neural network processing tiers ordered from a top to a bottom tier, each tier trained to generate conditioning information based on previous filter-bank samples and, for at least each tier but the top tier, also on the conditioning information from a tier higher up in the hierarchy, and an output stage trained to generate the probability distribution based on previous samples for one or more previous time slots and the conditioning information from the lowest processing tier.

Abstract: The present disclosure provides new variants of non-negative matrix factorization suitable for separating desired audio content from undesired audio content. In certain embodiments, a multi-dimensional non-negative representation of an audio signal is decomposed into desired content and undesired content by performing convolutional non-negative matrix factorization (CNMF) on multiple layers, each layer having a respective non-negative matrix representation. In certain embodiments, the desired content is represented by a first dictionary and the undesired content is represented by a second dictionary, and sparsity is imposed on activations of basic elements of the first or the second dictionary, wherein a degree of sparsity is controlled by setting a minimum number of components with significant activations of the first or second dictionary, respectively.

Abstract: The present disclosure provides new variants of non-negative matrix factorization suitable for separating desired audio content from undesired audio content. In certain embodiments, a multi-dimensional non-negative representation of an audio signal is decomposed into desired content and undesired content by performing convolutional non-negative matrix factorization (CNMF) on multiple layers, each layer having a respective non-negative matrix representation. In certain embodiments, the desired content is represented by a first dictionary and the undesired content is represented by a second dictionary, and sparsity is imposed on activations of basic elements of the first or the second dictionary, wherein a degree of sparsity is controlled by setting a minimum number of components with significant activations of the first or second dictionary, respectively.